Numerical Modeling of Water Blocks

[Highlander944]

Hey guys, it's been a while since I've posted... I've become gun shy and have been a bit busy.

Since some of my design methodology has been called into question I will not enter into a debate on certain areas of this topic, however I have some empirical information that is in direct contradiction to the graphs posted earlier.

In the early design of the Gemini block, I created a base that was approximately 2.5mm in thickness. I also created a block with the IDENTICAL water channel attributes, however the base was 1.25mm thick. In my testing, I only switched the blocks, nothing else was modified.

If I am to believe the model proposed here, my very very small flowrate (guessing, maybe 40-60gph) used in conjunction with the thick base should yield the best results... it absolutely did not. Infact the difference was 2 - 4C!!

My suspicion is that since we have a relatively small heat load (compared to the ability of copper and water to absorb the heat) that most of the thermal transfer takes place within a short distance of the core. Thus, the ideal block would probably be thin near the core and perhaps thicken by a 1mm or so as it extends outward.

One word of extreme caution. Most OCers out there are NOT mechanical. Copper is very very soft. If people use four springs with a combine compression force of 20+ lbs, you will begin to bend the bottom of the block if the block is thin! So keep in mind, when you design the block, that there are real machining and physical properties that should be considered.

I have a very slight design modification in mind to actually implement this theory and then test on a real system without any significant cost increase to the manufacturing process. Personally, I believe we are at the point of diminishing returns. 1C differences are purely bragging rights. =) The key factor which I believe MANY overlook is how much energy must be expended to obtain very good cooling.? The blocks designed for low flow allow the selection of a quieter, cooler running, more cost effective pump. Furthermore, the slower moving water can "sit" in a radiator or reservoir longer and perhaps enable better cooling of the water used in the system.

Ok fire away. =)

 

[Tecumseh]

Welcome back from your hiatus, Highlander944.

I think we are going to have these debates until someday
we have a theoretical model that matches measured data.
Thanks for sharing your tests with us.

Thin over the core and thickening toward the edges of the
block might increase the log-mean temperature difference
(LMTD) between the water and the block. This may explain
your results.

It sure would be nice if we could get a thermograph of your
block while running hot water thru it backwards.

Note: Before anyone has a heart attack over terminology,
LMTD really only applies to co and counter-flow heat exchangers.
For WBs maybe the term should be Integrated Mean Temp Diff
or Total Integrated Temp Diff.

 

[Aesik]

Lots to cover here, so bear with me please...

Some quick answers as to what I can/will include:

Variable base plate thickness: yes, it will be included and hopefully fairly soon. It's not an easy thing to do, but I should be able to set up the matrices that store the values to have null cells. It's just a question of making the program smart enough to know when a cell is null, and how to compensate for it. This isn't on priority list 1, but I'd rank it a 2.

Impingement cooling: unfortunately, this is an ENTIRELY different animal. This will not be added into this model at this time. I'm working on doing some research into how this has been most accurately modeled and deciding on how I will model it. For the first iteration of that project, I'm planning on doing a model of a jet onto a plane wall, then modify it from there. As was stated, impingement cooling is *extremely* difficult to model and there is not a solid method of doing so right now. Lately I've been considering going back to school to finish my doctorate, and if I do this is the project I would want to work on. Nasty, nasty stuff, but can be fun to work on if you have a good driving reason to do so.

Changing the flow direction: I'll have this implemented very soon. Last night I was working on the routine that will determine how much energy is dissipated along the path of the flow. I think I've come up with a satisfactory method to do so and am now working on it's implementation. Once I have that part pegged, I can recalculate the flow temperature at every node along the path. Not a simple thing, but not out of reach for the first version I'll release. Definitely a priority 1.

Highlander944: I'd really like to see exactly how you are doing your test setup to better understand how you are getting the temps you have. At this time, you can't realistically look at the preliminary graphs I've provided and apply them to your particular block unfortunately. The model is still in it's infancy and doesn't take into account your more complex geometry. What the model is good for is seeing how changing certain aspects of a design can change the performance, remembering that while all the graphs are pretty accurate, real world designs will take the graphs and shift them up/down/left/right.

BillA: I'm glad to see that you're interested in modeling the impingement cooling part of the problem. I'm with you on this one, I'd love to come up with a good way to do it. Unfortunately, as has been said, it is one of the most complex problems that engineers have ever been faced with. There are so many variables and so many things to take into account. My first pass at modeling it was a complete disaster. However, I'm more fired up than ever to get this problem licked. Well, I'm not expecting to do anything on the level of the Nobel Prize, but I would like to come up with a simulation that is at least somewhat helpful to our quest to design the next generation of cooling.

What I can predict right now is that impingement cooling would take my previously posted graphs, and scrunch them to the left. My meager calculations so far prove this, but I can't really get to the point of having meaningful graphs created yet.

Spartacus51: As I said above to Highlander944, this model can't be directly applied to your block and hoped to be perfectly accurate. Also note that while 10mm may seem the ideal thickness, going back up the graph to 5mm shows very little difference in temperature changes (at that particular flow rate). It's really a point of diminishing returns.

And finally...(if you're still with me!)...a few reminders and stats about my model. This is not a CFD model yet. I take general correlations of convective heat transfer to predict how the fluid will interact with the solid block. I am not predicting the actual flow profile, but rather it's interface with the solid. From there the finite difference model kicks in and predicts the temperature profile throughout the solid block, taking into account the material properties, simplistic geometry, effects of the flow and effects of the input power from the CPU.

I'm thankful for the interest in my little project. It's truly what keeps me going on working on something as masochistic and time consuming as this. I wish I had more time to work on it, but I'm relegated to my small amount of free time. Besides, it's hockey playoff time! A guy has to have his priorities! I usually park my butt and my laptop in front of the TV so as not to miss a minute of the Avalanche games while still getting some work done

Once again, thanks for all the support and the ideas, keep 'em coming.

 

[BillA]

Aesik
I am impressed by your appetite, for indeed you have a mouthful

for any (all ?) out there who do NOT have at least a MS involving Computational Fluid Dynamics, aka CFD;
I would like to recommend an introductory text:

"Using Computational Fluid Dynamics" by C.T. Shaw; 1992 Prentice Hall

most informative on all aspects

be cool

 

[Aesik]

And get comfy with Navier and Stokes, as they will now be your best friends and you will be spending a good chunk of your time with them

(sorry, couldn't resist the random CFD joke!)

 

[Les56]

In the early design of the Gemini block, I created a base that was approximately 2.5mm in thickness. I also created a block with the IDENTICAL water channel attributes, however the base was 1.25mm thick. In my testing, I only switched the blocks, nothing else was modified.

If I am to believe the model proposed here, my very very small flowrate (guessing, maybe 40-60gph) used in conjunction with the thick base should yield the best results... it absolutely did not. Infact the difference was 2 - 4C!!

. =)

I have less difficulty with these observations and the model than those of Billa .
That these results follow the high flow rate model I find acceptable. As I understand it, the predicted behaviour in the model at 50gph and 100gph only apply to the straight 1/4" sq pipe , maybe more complex designs change the the behaviour to that predicted at higher flow rates(eg 300gph) .The Gemini block is a spiral with DAI enhancement and probably showing more turbulence than the straight 1/4" sq pipe in the model.
That Billa(earlier in this thread) indicates higher temps at 300gph with thin bps(than thick, in unspecified wb) I find more difficult to reconcile with the model.This is surely low flow rate(eg 50gph) behaviour.

Confused as ever,
Les

 

[BillA]

don't feel alone Les, I'm right there with you

different setups, equipment, procedures, most with no semblance of calibration - or even correction
that the data is contradictory is a positive sign, how on earth could it possibly be the same (or similar ??)

my incomplete run was using a 462 (very different than a channel)
as I have sets of both Cu and Al bps, if I can ever get my hardware difficulties sorted out I'll run it correctly
am beginning to automate and integrate the system so it should be more accurate and faster - IF I can complete it

most of these 'problems' are related to the measurement increment being below the accuracy of the overall setup, indeed in many cases even below the resolution capability
- added to this is the system's (unquantified) period of variation, and the amplitude of same;
hence the results WILL be whatever the observer wishes them to be

comes with the territory

be cool

 

[Aesik]

I decided to post a 'pre-release' version of my numerical model in order to get some feedback on how to do the user interface. It's very crude right now and I'm mostly looking for suggestions on how it would be best improved for use by someone other than the programmer. Some feedback I'm looking for would be what types of units do people want to use as input (I'll probably make it possible to have several options), what kind of output would be useful, etc.

Remember that this is a model for a straight channel with the power input at the inlet side (it will be an option to change this in the first real release). I've hard coded the convergence criteria and the distance between nodes to be as accurate as possible while still having a fast running program.

This is a simple .exe (feel free to run a virus scan on it to feel safe!) and should be run from a command line. Just follow the prompts for information.

Thanks in advance for any feedback given. I'll post a list of future features when I get them all listed and prioritized.

 

[BillA]

won't run (4 tries)

units are fine
easy to load

but after 6 min or so shuts itself down (looping ?)
running on W2k on a Duron 750 with 1/2G RAM

be cool

 

[Aesik]

hrrmm....It runs in about 10 seconds for me using a 200mm channel length and 5mm thickness. This is on my PIII@980 and WinXP. I've also run it on a PIII 266 laptop with Win98 and it works fine too. Not sure what to say yet Bill, let's see if anyone else has any problems...

I guess I could put the iteration counter in there so it would print to screen what iteration it is on so you could at least see if it was running or if it just instantly locked.

 

[gone_fishin]

The program executes so fast on my machine that I don't have time to see the output. The command window shuts so you need a command to keep the window open at the end.

 

[Aesik]

Don't double click it to run it, start a DOS window and run it from there, that way the window won't close on you and you can repeat the run easily.

However, I will put a 'repeat' command in that will repeat or close the program for future use. Thanks.

 

[Les56]

Runs fine(2 or 3secs).Win98se on Celeron 600 but "performs an illegal operation" with channel lengths over 250mm.
I note the "taboo" Nusselt No. in the the output

 

[Aesik]

Added the option to repeatedly run, which will also fix the window closing problem for those that double-click the executable to open it. Expanded the matrix dimensions to allow for up 1250mm block lengths (is that long enough now Les? )

If this version won't work for you Bill, I'll stuff in an iteration counter for you to see if we can figure out the problem.

As far as the 'taboo' Nusselt number is concerned (do a forum search if any you don't know where this is coming from) it is a must in order to calculate the coefficient of convection. The only reason it prints out is because I personally want to see it (and needed it for de-bugging purposes) and it's easier to read it off the screen than to open the output file and read it there. Hopefully it won't cause any 'issues' =P

 

[Aesik]

Bleh...how come everytime I use the 'preview' button it wacks off my attachment? Here goes again....

 

[Les56]

Takes about 20-25sec on Celeron 600 to get the final Temp.
Easier to use than my "beloved " Kryotherm..
Gives a similar final answer(35.38c v 37c)[Edit :I made order of magnitude error ,Kryotherm temp should be 26c not 37c], to Kryotherm(straight U channel with "Forcer" mounted "near heat exchanger") :at least with parameters 400,300,100,6,6,6,and 200 . Intermediate"taboo" numbers differ. For Kryotherm the imput nunbers 6,6,6 refer to the wall thick(inc base), depth and width of the channel, I am unsure to what "the thickness of base" refers in your model.

 

[Aesik]

"Thickness of base" is just that, the thickness of the base itself. The depth and width of the channel are the internal dimensions of the channel itself. The depth of the channel is only used to calculate the flow conditions so far, not any heat transfer (that's where it starts to get really ugly, moving into the 3d realm).

What kind of Nusselt numbers are you getting out of Kryotherm? For that matter, where exactly in Kryotherm are you doing these calcs? I downloaded it before, but was extremely un-impressed with the user interface and the ability to find what you need in the program.

 

[Les56]

What kind of Nusselt numbers are you getting out of Kryotherm? For that matter, where exactly in Kryotherm are you doing these calcs? I downloaded it before, but was extremely un-impressed with the user interface and the ability to find what you need in the program.

I use the KryoTESC portion of the program.
The sub menu "Hot side parameters/Heat exchanger thermal resistance" to adjust the "heat exchanger" dimensions and material
The further sub-menu "Hot side parameters/Heat exchanger thermal resistance/Heat emission coefficient" to adjust the flow rates.
These two sub-menus can be operated independantly so that a "Charactaristic dimension" can be established in the "Heat emission coefficient" menu and used in different "heat exchanger" dimensions in attempt estimate the effect of different inlet nozzles.
I also,having first set the "Select Module" to a a 10mm sq. Peltier, use the "Hot side parameters/Intermediate thermal resistance" menu for the effect of thermal paste.

In the compared calculations with "Forcer mounted near heat exchanger" get :- Nusselt No 362 for water at 20c and 387 for water at 30c. No difference in "Heat excchanger thertmal resistance" prediction for two water temps.
However with "Forcer mounted on topof heat exchanger" get:- Nusselt No 208 and 222 with correspondimg increase of thermal resistance from 0.12 to 0.16 C/W[Edit : Should be 0.012 and 0.016 C/W] .

 

[Les56]

Oops .Silly me. Order of magnitude error in my Kryotherm predicted temps for 200mm pipe.
The "Heat Exchanger Thermal Resistances" should be 0.012c/w and 0.016c/w and not 0.12c/w and 0.16c/w.
This makes the Kryothem predicted temp 26c and not 37c which is vastly different to Aesik's model.
The intermediate Nusselt Nos are not effected as it was a stupid slip on my part.
Have/will edit previous posts.

 

[attack]

Hello, I'm new to OCer's forums but not new to OCing forums, you might have seen me at sharky's or Hardforums. I never came or really looked much at OCer's website and forums don't know why I just always thought I wouldn't find anything useful....anyway...

Thanks for all the hard work first of all:

That was the most interesting 3 page forum post i've ever seen! I myself have finaly gotten the oportunity to use a CnC machine and I want to make my own waterblock, 1/2 because I want to be original, 1/2 because I think I got some awesomes designs. So I ordered 4 blocks of aluminum and one block of copper. My plan was to mill out the four designs and put the best one on copper. Well I planned on making a base of 4mm with each block, testing then sanding down a mm then testing and do it one more time so i ahve tests at 4, 3, and 2 mm's.....

after reading all this work would have been in vain, I knew there was a difference between cu and al but never this big!

Well all my designs are on CAD on my other computer which doesn't have internet right now, I have burn them anyway so they can be brought to school and converted to mastercam for milling, but as soon as I get the CD burned I'll post screenshots of my designs...

one more thing, before the graphs showed around 10mm thickness for things, however, from the graph later showed that 2.5mm with 300gph and 4mm with 200gph was optimal, so should a 3.175mm base with a 250gph pump be about right? (I know there are a ton fo variables but would it be pretty close?)

Thanks for the input guys